Required questions: what forcing an answer costs you
Making every question required looks like a fix for missing data. The research says it trades blanks for dropouts and made-up answers. Here is when a required flag earns its place, and what to use instead.
By SurveyLane · The team building SurveyLane
Every survey builder has a required switch, and a complete dataset feels like a clean one. A required question still cannot make anyone answer. What it does is leave the person who wanted to skip with a choice: answer after all, make something up, or close the tab. Only the first shows up in your export as what it is.
What the required switch actually does
When a question is required, the respondent cannot move on until it has an answer. Methodologists call this forced answering, forced response or a hard reminder. It is a design setting, so it works on any question format, open text fields included.
The setting only acts on the person who tried to skip. So ask what that person does when you stop them. They can give an honest answer after all. They can type or tick something just to get past the block. Or they can leave. A required flag only pays off if the first group is much bigger than the other two.
Why people leave a question blank
People rarely skip out of laziness. Usually the question gives them no way to answer truthfully:
- They do not understand the question, or read it two ways.
- None of the options fit their situation.
- They do not know the answer, and guessing feels wrong.
- The question is personal and they would rather not share.
- The question does not apply to them, and the routing missed it.
A cluster of blanks on one question tells you the wording or the answer list is broken. Force an answer and that signal turns into noise you cannot filter out, because you no longer know which answers came from people who would have skipped.
The post on multiple-choice questions covers the answer-list side, including why "don't know" and "not applicable" are different answers.
The evidence: forced answers push people out
The clearest experiment is by Jean Philippe Décieux, Alexandra Mergener, Kristina Neufang and Philipp Sischka, published in Psihologija in 2015. They emailed about 20,000 students at two German universities, and 625 took part. The questionnaire had roughly 100 items about relationships and sexuality, so plenty of personal questions. Respondents were randomly assigned to one of three versions: skip anything you like, answer everything, or answer everything with a "prefer not to answer" option in each list.
In the forced version, 80 of 205 respondents (39 percent) tried to skip at least once and hit the error page. Of that group, 35 percent dropped out. Among everyone who was never blocked, and in the version with "prefer not to answer", dropout was 9 percent.
Look at where they left. In the forced group, dropout jumped by 7.5 percentage points at an open question about which traits of their partner respondents would like to change, and by 8.7 points at a question about their own experience with sexually transmitted diseases. These people were near the end. The authors call them "bad dropouts": motivated respondents you lose because of the design.
Décieux and colleagues also cite a 2007 study by Stefan Stieger, Ulf-Dietrich Reips and Martin Voracek that found 64.1 percent dropout in a forced-answer group, against 18.6 percent when skipping was allowed.
Panels behave differently
Some studies find no dropout penalty at all. Gerald Albaum and colleagues in 2010, and Catherine Roster and colleagues in 2014 with samples in the United States and Hong Kong, found that forced answering did not raise dropout and did cut item nonresponse significantly. Both used online panels.
Panel members are paid per complete and used to finishing questionnaires, so one blocked question rarely makes them quit. Your customers and employees owe you nothing, and the question a panel member shrugs at can be the one where a customer gives up.
If you buy panel sample, a required flag costs you few completes, though it may still cost you honesty. If the people you survey are doing you a favour, expect the student result.
A forced answer is not always a real answer
Dropout is the cost you can count. The second cost is harder to see. At the end of the Décieux survey, respondents were asked, in neutral wording, how many questions they had not answered honestly. Among those who had been blocked, 25 percent admitted to at least one. In the version with "prefer not to answer" it was 13 percent. And the most irritated people in the forced group had already quit before they reached that question, so the real gap may be larger.
Psychologists call this reactance: take away someone's freedom to decline and some push back, by quitting or by giving an answer that is not true. Picture a required income question in a customer survey. Someone who does not want to share ticks the first bracket, or the middle one. The column in your export is complete, and your income breakdown now contains answers nobody meant.
Knowledge works the same way. Ask a required "how often did you contact support last year?" and people who do not know type a guess, usually a round number. The post on asking sensitive questions explains why people shade their answers in the first place. Forcing a response adds to it.
"Prefer not to answer" and its price
A "prefer not to answer" option looks like the obvious compromise. The question stays required, so nobody skips by accident, and people still get an honest way out. In the Décieux study it removed the dropout penalty, and self-reported faking was about half that of respondents who had been blocked.
Edith de Leeuw, Joop Hox and Anja Boevé found in 2015 that offering such an option explicitly increases missing data and often lowers the reliability of the answers you do get. An easy exit gets taken.
So offer it where declining is a legitimate answer. Income, health, sexuality, religion, a question about a named manager: here "I'd rather not say" is information, and you should accept it. A question about which feature someone used most last week does not need one. If people cannot answer that, add a "none of these" option or fix the routing.
In the analysis, count these answers as missing, not as a point on your scale.
Soft prompts: ask once, then let go
There is a better default than either extreme. Let people skip, and when they do, show a short message asking whether they meant to. Then let them continue whatever they decide. This is a soft prompt, or soft reminder.
The evidence for it is consistent. As Décieux and colleagues summarise, DeRouvray and Mick Couper in 2002 and Stéphane Ganassali in 2008 both found that a warning on a skipped question reduced item nonresponse, even though leaving it blank was still allowed. Tarek Al Baghal and Peter Lynn compared a prompt straight after the skipped question with a reminder at the end of the questionnaire. The direct prompt worked much better and brought item nonresponse down close to face-to-face levels. In the de Leeuw study, allowing the skip and following up with a polite probe gave the highest response of any design they tested.
A soft prompt catches the accidental skip and nudges the mildly reluctant. The person with a real reason to decline is left alone. The wording decides whether it feels like a question or a wall. Compare:
- "Error: please answer all questions before continuing."
- "You haven't answered this one. It helps us understand why customers cancel, but you can skip it if you prefer."
The second says why the answer matters and that skipping is fine. Put it next to the question and show it once. A prompt that fires on every press of next is a hard reminder with extra steps.
Which questions earn a required flag
Some questions really do need an answer, because the survey breaks without it. The test: if this is blank, does the rest of the survey still work?
- Consent questions, where you cannot legally or ethically continue without a yes.
- Screening questions that decide whether someone belongs in the sample.
- Routing questions whose answer decides which block comes next. The post on conditional logic shows how a branch hangs on that answer.
Your key outcome question is missing from that list on purpose. It is tempting to require the NPS question or the overall satisfaction score, since that is the number you report. But a forced score from someone who did not want to give one is the answer you can least trust. Use a soft prompt and report how many people answered.
Keep those few factual and impossible to misread, so nobody wants to skip them.
Never force an open answer
A required open text field collects every problem above. Typing takes effort, especially on a phone, and someone with nothing to say has to invent something. You get "na", "nothing", "...", a single letter. The Décieux data showed one of its largest dropout jumps at exactly this kind of question.
Make open questions optional and put the one you care about most early. The post on mobile survey design covers why typing on glass shortens answers, and the guide to analysing open answers deals with the text you do get.
Reading the blanks you allowed
Report the base for every figure. "62 percent satisfied (n = 412)" shows the reader that fewer people answered this question than took the survey. Never treat a blank as zero or as the lowest scale point, because that turns a non-answer into a strong opinion. And watch out for analyses that quietly drop every respondent with any blank. Combine ten questions, remove everyone who skipped one, and you may lose a third of your sample. Not a random third, either.
Skip rate per question is also a quality signal. A question skipped five times as often as its neighbours is badly worded, or sensitive, or missing an option. The post on spotting bad survey responses covers the other warning signs. With the SurveyLane MCP server you can ask an assistant which questions have the highest skip rate, and whether the people who skipped differ from those who answered on everything else.
Test the decision before launch
You cannot guess which questions people will want to skip. Soft-launch to a small slice of your sample with skipping allowed everywhere, and the skip rate per question shows you where the wording needs work. The guide to pretesting a survey explains how to set that up.
Frequently asked questions
Should every question in my survey be required?
No. Forcing every answer raises dropout among people who wanted to skip and pushes some of the others into answers that are not true. Require only the questions the survey cannot run without, such as consent, screening and routing questions, and use a soft prompt everywhere else.
Does a required question improve data quality?
It improves completeness, which is a different thing. You get fewer blanks, but some of the answers filling them were given only to get past the block. In one experiment, respondents who had been blocked admitted to dishonest answers almost twice as often as those who were offered a way to decline.
What is the difference between a soft prompt and a required question?
A required question will not let the respondent continue until it has an answer. A soft prompt shows a short message when someone skips, asks whether they meant to, and lets them go on either way. Soft prompts cut blanks without the dropout penalty of forcing.
Should I add a prefer not to answer option to every question?
Only where declining is a legitimate answer, such as income, health or other personal topics. On ordinary questions it hands people an easy exit and can lower the reliability of your data, so fix the answer list or the routing instead.
